2.1. Device Characteristics
Figure 1 presents the device properties of the 1:1.5 by weight PTB7:PC71BM blend devices (chemical structures in Figure 1c) cast from chlorobenzene (CB) solutions without or with 3 volume% DIO as additive (this point forward labeled “CB” and “CB+DIO” samples). Both device performance (Figure 1a) and external quantum efficiency (EQE, Figure 1b) improve drastically with the addition of DIO, similar to the original study on this effect.2 In fact, the EQE and device parameters as outlined in Table 1 are nearly identical to those reported previously2– especially to the CB+DIO device where only the fill factor is different and fully accounts for the difference in efficiency between the two studies. The lower fill factors in our devices may result from unoptimised contact to electrodes that reduce the internal electric field in the device.3 The major differences between the two samples in this study are a doubling of the short circuit current and a relatively modest 19% gain in the fill factor resulting in a more than doubling of the efficiency.
Figure 1. Device and material characterization: (a) Current density vs. applied voltage of the PTB7:PC71BM devices under AM1.5 illumination, (b) external quantum efficiency, (c) material chemical structures, and (d) difference in quantum efficiency between the two devices superimposed by a fit from a linear combination of UV–visible absorption spectra from pristine films of each molecule.
Download figure to PowerPoint
Table 1. Device parameters: PCE = JSC*VOC*FF/I0, where incident intensity I0 = 100 W m−2.
|Sample||JSC [mA/cm2]||VOC [V]||FF [%]||PCE [%]|
To investigate the source of this gain in solar harvesting, the differential EQE spectrum (EQECB+DIO–EQECB) was fit to a linear superposition of the pure UV-visible absorbance for each component material and is presented in Figure 1d. This simple analysis allows for a rough calculation of the relative gain in efficiency originating from absorption by the two molecular species. Interference effects in the device were simulated using prior established methods,19–22 but did not appear to affect this analysis substantially (see Supporting Information for details). The simulations also showed that reorganization of the lateral morphology would not affect the absorption of the two species, indicating that any gains in the EQE must be from increased exciton dissociation, separation, or charge transport. The resulting relative efficiency gains of 42% for the polymer versus 58% for the fullerene show that the increased efficiency is preferentially from charges originating from PC71BM absorption. This is an important result as most morphological optimizations in the past have focused on the contribution of the polymer and suggests a closer look at fullerene morphology is also warranted in these systems.
2.2. X-Ray Diffraction
As a change in crystallinity has been linked to device performance in multiple material systems, we probed this aspect using grazing incidence wide angle X-ray scattering (GIWAXS) at beamline 7.3.3 of the ALS23 with primary results presented in Figure 2. The 2D data in Figure 2a,b, as well as the profiles in Figure 2c,d, reveal patterns very similar to those reported previously for these films.14, 15, 24 The variation of the intensities of the polymer reflections with azimuth (shown in the sector profiles of Figure 2c and 2d) confirm the preferential orientation of the polymer aromatic planes of the crystallites is ‘face-on’ with the substrate and occurs in roughly equal levels in the two samples.
Figure 2. Grazing incidence X-ray scattering study: Panels (a), and (b) show background subtracted CCD output of the CB and CB+DIO samples, respectively. (c) and (d) are ±10° sector profiles along the directions specified in the legend and circularly averaged profiles from (a) and (b), respectively. The sharp dip in Z-profiles at ∼6 nm−1 is from a background subtraction artifact. (e) Circularly averaged profiles of the two samples compared directly. Detailed analysis is described in the Supporting Information.
Download figure to PowerPoint
Figure 2e compares the circularly averaged profiles of the two samples. The plots overlap nearly perfectly, demonstrating that the crystallinity of the film changes very little when DIO is added to the casting solution. The only significant difference is in the position of the PTB7 (100) reflection, which represents the lamellar spacing. A decreased peak position indicates that the lamellar spacing increased when DIO is added to the solution. The profiles were fit to a series of Lorentz functions on a double exponential background (see Supporting Information) and confirm that the spacing changes from 17.1(1) Å without DIO to 18.9(1) Å with the additive. No other change of d-spacing was detected, and all other d-spacing values of polymer and fullerene (see Table S1) correspond quantitatively with those reported earlier.14, 15, 24 It is difficult to hypothesize any major effect to devices from the change in lamellae spacing measured here, since charge generation or transport is not thought to occur in this direction of a polymer crystal.
To assess the level of crystallinity, the peak area and coherence (2π/FWHM) – which roughly corresponds to crystallite population and size, respectively – were analyzed from the fits. No significant difference between the two samples was detected for both parameters of all analyzed peaks (see Figure S2). In the polymer, the PTB7 (100) peak area is substantial but has a coherence of only 30 Å, resulting in less than two lamellar spacings per ‘crystallite’. For the PTB7 (010) π-stacking reflection – the direction corresponding to charge transport – the measured coherence results in approximately five stacked molecules but with the fitted values of coherence and peak area agreeing with zero due to the surrounding fullerene reflections (See Figure S2). All fullerene peaks result in packing coherences of ∼20 Å or less, but with a measured (100) d-spacing of 9.7(1) Å, this only corresponds to a ‘crystal’ of a few molecules. This coherence length is typical of most polymer/fullerene blends and has been used to postulate a lack of fullerene crystallinity in these films.25 These results agree with previous studies demonstrating that the crystallite population is very small.14
2.3. Resonant Microscopy and Thermodynamic Miscibility
Since the crystallinity is very low and does not change between the samples, it cannot be the reason for the dramatic change in device performance. Therefore, the overall morphology and composition was investigated using spectromicroscopy with a scanning transmission X-ray microscope (STXM) at the 184.108.40.206 beamline of the Advanced Light Source.26 Due to diverse molecular resonances of the two components, an image of a film acquired at multiple X-ray energies can be quantified into composition at each pixel via knowledge of each material's mass absorption spectra as calculated from near edge X-ray absorption fine structure (NEXAFS) measurements (see Figure 3a).27–30 Presented in Figure 3, the composition maps of the two films reveal extremely different morphologies with ∼200 nm sized fullerene dispersions in a mixed matrix for the CB samples compared to a much finer texture in the CB+DIO sample, in agreement with AFM data (Figure S4) and EFTEM tomography.14 The histogram and line profiles in Figure 3d,e show that the matrix in the CB sample is 30(2) wt.% fullerene whereas the fullerene dispersions are pure to within an uncertainty of 2 wt.%, similar to the domain compositions of MDMO-PPV/PC61BM blends.28 Although the theoretical resolution limit of the microscope under ideal operating conditions is 31 nm, quantitative measurements become unreliable in features far larger due to the point spread function of the X-ray beam artificially reducing the measured purity of the domains.30 (See the round-topped fullerene dispersions in Figure 3e, for example.) The features seen here are at the limit of accurate compositional analysis, and this is the first time the domain composition of device active layers has been measured in such a high performing material system. By comparison, the CB+DIO sample is either nearly fully mixed or has domains that are smaller than the resolution limit of the X-ray microscope. The composition analysis on this sample, however, cannot be conclusive as the size of the morphology imaged with STXM is far below the resolution limit for quantitative analysis.
Figure 3. Resonant microscopy: (a) NEXAFS spectra of the component materials demonstrating the nature of the contrast. (b) & (c) Film composition maps (same magnification) of the CB and CB+DIO films, respectively, obtained from imaging at energies 284.4 eV, 320 eV and 350 eV in (a). (See Supporting Information for details.) (d) Histograms of (b) & (c). (e) Line profiles taken from the yellow lines in (b) & (c). The black dashed lines in (d) & (e) represent the measured composition of the matrix phase in the CB sample.
Download figure to PowerPoint
To assess the role that thermodynamics plays in the composition of the domains, the molecular miscibility was measured as done previously for P3HT7 following the detailed analysis outlined elsewhere.30 Figure 4 presents fitted NEXAFS spectra for blend films as cast and annealed (at 140 °C) to compositional equilibrium. The annealed films developed many-micron sized PC71BM crystals as seen in other studies,7, 12, 13 which gradually depleted the surrounding film. The statistical average miscibility from multiple measurements resulted in 29(1) wt.% PC71BM in PTB7 – the same value as measured in the matrix of the as-cast CB films. This result suggests that the initial fullerene composition of 60 wt.% in solution is high enough for spontaneous phase separation (above the binary spinodal) and there is sufficient time and mobility during solvent evaporation for the matrix to reach the thermodynamic limit, which should in all cases be a lower bound for the fullerene concentration as the solvent concentration diminishes to zero. A miscibility this high may be important for electron transport in the matrix as it is well above the percolation threshold of ∼20 wt.% demonstrated recently by Vakhshouri et al. in electron transport measurements through PC61BM in regiorandom P3HT.31
Figure 4. PTB7 miscibility with PC71BM: Panel (a) and (b) show composition measurement of as cast blend film and one annealed at 140 °C for 95 hours in an inert environment, respectively. Red lines/symbols are NEXAFS data/uncertainty, black lines are fits, and grey lines are residuals dividedbytheuncertainty. Yellow and blue lines are PTB7 and PC71BM reference spectra used in the fit and plotted on a separate Y-axis for clarity.
Download figure to PowerPoint
2.3. Resonant Scattering
To probe morphology below the resolution limit of X-ray spectromicroscopy, we conducted R-SoXS experiments on the same two films investigated with diffraction and microscopy. Scattering intensity arises from structure in the film through the contrast function defined as C = |Δδ2 + Δβ2|E4 = Δn2E4 where n = 1 − δ + iβ is the materials' complex index of refraction and E is the photon energy. Displayed in Figure 5a, this function can be calculated from the NEXAFS measurements (Figure 3a).32 There are two primary ways that structure in the film can cause scattering: roughness/thickness variations through the vacuum contrast (blue and red dashed lines Figure 5a), and material domains (black line Figure 5a). To probe material domains, we conducted the scattering experiment at 282.5 eV where the vacuum contrast of both components are minimized relative to the material contrast (green arrow Figure 5a). Notably, using this energy takes advantage of the materials' phase (i.e. δ) contrast rather than absorption (β) contrast used in STXM and thus largely avoids the associated radiation damage effects on the sample and fluorescent background that occurs when core-holes are created.
Figure 5. R-SoXS study: (a) Contrast functions (Δn2E4) for each material with vacuum (n = 1) and between each material calculated from NEXAFS measurements in Figure 3a. The assumed density used in the calculation was 1.12 and 1.25 g/cm3 for PTB7 and PC71BM, respectively based on similar materials.46, 47 Green arrow indicates where data was acquired. (b) Azimuthally integrated scattering profiles with associated peak fits and calculation of the Total Scattered Intensity (TSI). (c) Normalized histogram of scattering intensity versus fullerene domain size from calculations described in the text. Extra scattering from interfaces in the CB sample (red curve) below ∼80 nm likely spuriously enhances the apparent domain population at this size scale.
Download figure to PowerPoint
The results of the R-SoXS experiment are presented in Figure 5b, each film showing a well-defined broad peak, which were each successfully fit to Voigt peak functions with a log(q) axis (a second small peak for the shoulder in the CB sample). Due to the large polydispersity of the domain sizes as evident in the STXM images, any form factor from the dispersions themselves will be washed out, leaving only the structure factor information.33 These peaks, therefore, represent domain spacing at d = 2π/q (see top axis of Figure 4b), which is reduced considerably in the presence of the DIO additive. Particularly important to characterizing this system was the use of the proper X-ray energy in reducing vacuum contrast. Scattering profiles acquired at 272 eV, where vacuum contrast dominates, show a double peak in the CB+DIO sample (Figure S6), which could erroneously be interpreted as a hierarchical structure or a morphology with multiple lengthscales. This highlights that a judicious choice of photon energy is necessary in resonant scattering.
The scattering profiles can be processed further to extract the domain purity of an assumed two phase system through the total scattering intensity (TSI) as first shown by Porod and therefore also referred to as Porod's invariant.33–35
where Δρ12 is the difference in electron density between the two phases, vi is the volume fraction of each domain and V is the total illuminated volume. This quantity is insensitive to morphology as domains of different sizes scattering to different angles (q). What does affect the TSI is domain contrast, which is therefore an important measure of the domain purity. We extend equation 1 for resonant X-rays through the following conversion:
where f1-if2 are the complex elemental scattering factors, nj are the number densities for the jth element in each component molecule, and α is a factor containing only universal constants (see Supporting Information). The left equality in equation 2 comes from the fact that the scattering factor is equivalent to the effective number of electrons that scatter, while the middle equality converts from atomic view to that of a molecule. The final equality converts from scattering factors to optical constants. Thus
where the subscripts indicate contrast between domains (not materials). If the materials are mixed within the domains, this contrast will be reduced accordingly, and so will the TSI.
As shown in the legend of Figure 5b, the TSIs from the two samples investigated differ by only 12% as calculated from equation 3. The illuminated volume V was measured to be the same for both samples (see Supporting Information). As Δn scales with purity and the volume fraction of the domains has to be very similar, this difference in the TSI means an approximately difference in relative phase purity between the two samples. Remarkably, this implies that the relative domain composition is nearly identical in both samples and only the size distribution has changed.
The nature of the domain interface can also be probed using Porod's law for which the high-q limit of the scattering profile follows a power law.33, 35, 36 Smooth, sharp interfaces result in an intensity fall off of q−4. Fractal interfaces fall off slower (the effective interface area grows as smaller length scales are probed increasing scattering), while diffuse interfaces fall off faster (more mixing produces less scattering). This analysis was possible on the CB sample, where fits in the Porod region resulted in a power of ∼ -4.5 (Figure S5), suggesting diffuse interfaces–in contrast to analysis in P3HT:PC61BM where the power-law scaling with q is less than -4,10 and thus the interfaces in P3HT:PC61BM are “fractal” due to crystallite boundaries. The small quantitative reduction of the TSI measured here between the samples is possibly due to this interface diffusiveness, where the much smaller domains in the CB+DIO sample have a higher interface-to-volume ratio, and therefore the average purity over the entire domain would appear to decrease slightly. An alternative explanation may be that, although the fullerene dispersions are also pure in the CB+DIO sample, the matrix is slightly enriched in PC71BM which is held in solution longer by the DIO and eventually trapped in the frozen PTB7-rich matrix.
Having established that the domain compositions for both samples are essentially equal, the intensity profiles can be inverted into real space histograms of fullerene domain diameter (conversion from spacing to diameter in supporting information). It is important to note that this simple inversion can only be accomplished with independent confirmation that domain form factors do not contribute to the scattering as is the case here. These scattering intensity histograms, representing the volume distribution of domain sizes, are plotted in Figure 5c revealing a monomodal distribution rather than bimodal or hierarchical. The difference in domain size distribution between the samples is stark, with a dominant domain size of 177 nm in the CB sample reduced to that of 34 nm when DIO is added. The latter size is far more amenable to efficient dissociation of excitons created in the fullerene agglomerates at a domain interface and is too small to be imaged by the STXM, explaining the results in Figure 3c. This distribution is fairly wide, however, with some domain diameters well below 10 nm while others are 100s of nm in size. A tighter distribution around 10 nm may improve exciton capture of the PCBM further.